100k nquads would be very very fast, probably not even adequate enough to benchmark. The live loader would insert a pre-formatted version of this in probably 1-2s maybe (big swag but you get the point). Obviously it matters what your dgraph is provisioned with but assuming an appropriately sized system, it will be super quick.
Using upserts to idempotently insert each of 100k things to the same node (xid->uid translation) would probably add a small amount, but reading 100k strings out of dgraph is real fast, would probably add only another second or so? Again the numbers here are so small you would not get a consistent time on execution.
But it all depends what data shape you have, what indicies are being built when you insert each thing (like trigram index with long strings can increase the amount of data you are actually saving by a lot).
Sorry, not a real answer other than ‘probably pretty fast’